I know how it works :p
#OBOS (not vibecoded)
dense vigil
You can have a program interpret the noises
So when you play a diffrent note it does a different action
sterile sparrow
real pecan
oberrow is the younger version of dwl
flint idol
oberrow is the younger version of dwl
thats the guy on the cherno server who deleted his account right
dreaming warlord or smth
real pecan
lol yeah but hes back
flint idol
ah
torpid wigeon
thats the guy on the cherno server who deleted his account right
the cherno 💀
dense vigil
What’s the cherno
real pecan
C++/game YouTuber
real pecan
Probably
dense vigil
I’ve long left tutorial hell though thankfully
torpid wigeon
What’s the cherno
he makes/made c++ tutorials
i don't like his videos
he yaps for 40 minutes about a topic that you learn within 10 days of writing C++ and is just a common skill issue
dense vigil
yeah i remember him being ass lol
stark sigil
somewhat (un)related but i wonder how people dont understand variables in programming languages
its like math variables
torpid wigeon
what
stark sigil
what
idk some beginners have problems with them for some reason
sure, fine with the uint's and ints and other types
but the variables themselves idk
torpid wigeon
ive never met such a person
real pecan
he yaps for 40 minutes about a topic that you learn within 10 days of writing C+...
His biggest thing was game engine series
lean glen
its like math variables
it's not
torpid wigeon
me when const variables in math
lean glen
also x = x + 1 would probably give a heart attack to a mathematician
real pecan
also x = x + 1 would probably give a heart attack to a mathematician
Cc @barren merlin 
stark sigil
also x = x + 1 would probably give a heart attack to a mathematician
x += 1
barren merlin
also x = x + 1 would probably give a heart attack to a mathematician
Eh
They would say the set of solutions is the empty set
tall sluice
yeah i remember him being ass lol
I learnt programming from him
J feel like he used a lot of time explaining things that could be explained in half the time
uncut narwhal
@flint idol btw how does obos handle handles? like managing them, sharing if thats a thing you do, associating the number with an object, etc
flint idol
ill explain after dinner
flint idol
yeah sorry I forgot to answer
ill answer tomorrow or you can read src code
src oboskrnl handle.h/c
flint idol
oki
uncut narwhal
(mostly looking for inspo at this point, ive learned semi-cursed things about zig std hash maps that make them goated for all sorts of weird map-like structures and im sure im going to be using something like that for my nonsense)
flint idol
tomorrow
I might do a proper devfs
like a mountable one
I had an idea today
for how to do it
slow dove
write it down
flint idol
rn I have magic functions to register dirents in /dev
but
and the goal is to change them in some way that makes it not just register in /dev
through some sort of literal devfs driver
the idea hasn't fully had the time to
develop
and consolidate
hopefully doing this will help me hold steady
ive been going in and out of sanity over the past month because of <redacted>
I might even implement another congestion control algorithm for obos
and in theory after I do that I can merge the tcp congestion control branch
and if after that im feeling fine I can fix the xhci driver
ive had some theories on why it's broken
while im doing that I also might properly implement tcp SACK
on obos
and by properly implement
i simply mean
fix bugs with it
and i might also try to optimize the tcp stack a bit more
and hopefully I will follow through on this instead of bedrotting and being sad
sometime between this
I might try to fix a longstanding bug in the ps2 driver
flint idol
ive had some theories on why it's broken
if only I had theories on why im broken 
jk im 50% fine
50% is heartbroken
uncut narwhal
🪦
uncut narwhal
src oboskrnl handle.h/c
oh btw your handle source was very clear and understandable, no further questions at this time
torpid wigeon
I might do a proper devfs
devfs is just a tmpfs singleton
blazing python
wait until he finds out I dont have tmpfs
what filesystems you have now
flint idol
I have tmpfs but only for initrd
and ext2
which has a 90% chance of corrupting the filesystem if you create a new file
uncut narwhal
yknow that sounds like the sort of thing you should fix lmao
dense vigil
yknow that sounds like the sort of thing you should fix lmao
Nah I love playing Russian roulette with my file system
flint idol
talk cheap 
uncut narwhal
lmao
flint idol
which has a 90% chance of corrupting the filesystem if you create a new file
which may also be in part due to the fact that the kernel might crash while creating the file leaving changes half-synced to disk
uncut narwhal
yeah thatll do it
flint idol
do people actually listen to malcolm todd
uncut narwhal
idk who that is
flint idol
fine shyt told me thats her favorite artists
uncut narwhal
oh if musician thats why
i listen to this one band from the 70s and thats about it
flint idol
I see
uncut narwhal
oh and also listen to crane wives
flint idol
ur like unc right
uncut narwhal
but those are the two
flint idol
uncut narwhal
ur like unc right
more like aunt but yeah
im only 27 but like im into the older music generally
uncut narwhal
if unc is gender neutral then yeah
flint idol
if i called my sister unc she wouldn't care
uncut narwhal
honestly ive got more unc vibes than aunt vibes despite being a woman
flint idol
ok one of these songs is maybe good
uncut narwhal
yknow tbh its probably all the handwriting html thats giving me so much unc vibes
flint idol
yeah
flint idol
ok one of these songs is maybe good
yes for the most part this shit is ass
uncut narwhal
https://www.youtube.com/watch?v=jRwmkZ49jVM this is the shit i listen to lmao
flint idol
lol
uncut narwhal
making me think, does obos have audio drivers yet actually?
flint idol
yes
uncut narwhal
cool
uncut narwhal
slick
flint idol
cc @white mulch
I haven't been able to get it to work on real hardware but at the same time my setup could just be wrong
and I haven't been able to get it to work on vbox
uncut narwhal
i wonder if i can straight pass through my hd audio device to qemu now
real pecan
oh idk how windows pci passthrough works
on linux u just unbind the device from the normal driver and bind it to vfio i think
uncut narwhal
lmao
white mulch
does uhda work on real hw?
depends on the hw, it doesn't really have any quirks implemented except for not using msi on nvidia cards (though iirc most of the quirks have to do with the extra functionality and not so much with basic playback except at least some dma position reporting thing)
real pecan
ah cool
white mulch
but at least on my laptop that has the intel smartsound technology thing it works (+ at least with passthrough on my pc too)
uncut narwhal
so sounds like a fairly standard cid/hid moment where the basic functionality on the cid has fairly few quirks and if you want extra identify based on vendor and devid and shit to get specific with it
weak kestrel
more mouse bites
flint idol
I feel like thats a reference I should get considering im on season 6
btw is season 6 after a month of watching the show a bad thing
weak kestrel
btw is season 6 after a month of watching the show a bad thing
considering I finished 7 season star trek show in like 3 weeks, idk
flint idol
https://www.youtube.com/watch?v=ry-Z6_yfZoE
lmfao
ornate ginkgo
considering I finished 7 season star trek show in like 3 weeks, idk
Which one?
weak kestrel
Which one?
all of them 
inland radish
yay!
blazing python
tmpfs is just in-memory only fs, yeah?
yes
blazing python
im implementing a proper tmpfs
do tree tmpfs
uncut narwhal
yes
cool. im a general hater of devfs and sysfs so i always forget how the other in-memory fses work
flint idol
do tree tmpfs
I dont think there are many other alternatives
trim bay
tmpfs is just in-memory only fs, yeah?
yeah, the usual implementation is that you never evict any data from the page cache and hold an extra reference to every namecache entry until that entry is unlinked
so it still has full POSIX semantics unlike procfs and sysfs
uncut narwhal
yeah, the usual implementation is that you never evict any data from the page ca...
so i dont know anything about namecache but the rest makes sense
tall sluice
@flint idol can U explain now?
So like
destdir is just where the package gets installed to
And
Bin pkg dir is where the pkg gets created to?
flint idol
yes
flint idol
also I misunderstood this and thought u meant u would explain lmao
tall sluice
also I misunderstood this and thought u meant u would explain lmao
ohhhh
no
mb
xD
flint idol
in hindsight it made a lot more sense for that to have been a typo
tall sluice
in hindsight it made a lot more sense for that to have been a typo
it was a typo 😔
flint idol
xhci today 
weak kestrel
damn everyone is doing nice stuff
I gotta step my game up
but that's for the future me to worry about
flint idol
i implemented some dumb MMIO_WRITEn and MMIO_READn macros
#if __x86_64__ || __x86__
# define MMIO_WRITE8(address, value)\
asm volatile ("movb %1, %%al;\
movb %%al, (%0)"\
::"r"(address),"r"((uint8_t)value)\
:"%eax","memory")
# define MMIO_WRITE16(address, value)\
asm volatile ("movw %1, %%ax;\
movw %%ax, (%0)"\
::"r"(address),"r"((uint16_t)value)\
:"%eax","memory")
# define MMIO_WRITE32(address, value)\
asm volatile ("movl %1, %%eax;\
movl %%eax, (%0)"\
::"r"(address),"r"((uint32_t)value)\
:"%eax","memory")
#if __x86_64__
# define MMIO_WRITE64(address, value)\
asm volatile ("movq %1, %%rax;\
movq %%rax, (%0)"\
::"r"(address),"r"((uint64_t)value)\
:"%rax","memory")
# endif /* __x86_64__ */
# define MMIO_READ8(address)\
({\
uint8_t _result = 0;\
asm volatile ("movb (%1), %%al;\
movb %0, %%al"\
:"=r"(_result):"r"(address)\
:"%eax","memory")\
})
# define MMIO_READ16(address)\
({\
uint16_t _result = 0;\
asm volatile ("movw (%1), %%ax;\
movw %0, %%ax"\
:"=r"(_result):"r"(address)\
:"%eax","memory")\
})
# define MMIO_READ32(address)\
({\
uint32_t _result = 0;\
asm volatile ("movl (%1), %%eax;\
movl %0, %%eax"\
:"=r"(_result):"r"(address)\
:"%eax","memory")\
})
# define MMIO_READ64(address)\
({\
uint64_t _result = 0;\
asm volatile ("movq (%1), %%rax;\
movq %0, %%rax"\
:"=r"(_result):"r"(address)\
:"%eax","memory")\
})
#endif /* __x86_64__ || __x86__ */```Why not a static inline function here?
Then set [[gnu::alwaysinline]]
uncut narwhal
```c
#if __x86_64__ || __x86__
# define MMIO_WRITE8(address, value)\
asm v...
couldn't you just constrain value to use the a register?
ruby condorBOT
flint idol
because i love macro hell
dense vigil
Pls no macro hell
uncut narwhal
lol
flint idol
couldn't you just constrain value to use the a register?
idk how inline assembly works
flint idol
i dont even make myself read it
dense vigil
idk how inline assembly works
You can set register constraints yes
uncut narwhal
you make the constraint "{al}" or "{ax}" or whatever
instead of r
and then the compiler guarantees goes in that register
uncut narwhal
thats what i do for my mmio read/write helpers
dense vigil
This is how you have to do it in C
dense vigil
thats what i do for my mmio read/write helpers
I need to not use volatile C loads and stores
I mean I already have it abstracted to functions
So it’s like a very fast fix
uncut narwhal
I need to not use volatile C loads and stores
volatile loads/stores would maybe be ok if not for the fact that theres a brief range of amd processors that require it to be using the A register for some fuckin reason
dense vigil
volatile loads/stores would maybe be ok if not for the fact that theres a brief ...
Yup
flint idol
so this:
asm volatile ("movb %1, %%al;\
movb %%al, (%0)"\
::"r"(address),"r"((uint8_t)value)\
:"%eax","memory")```
becomes this?
```c
asm volatile ("movb %%al, (%0)"\
::"r"(address)\
:"%a"(value),"memory")```And it has to be mov
blazing python
```c
#if __x86_64__ || __x86__
# define MMIO_WRITE8(address, value)\
asm v...
is this because of xhci
dense vigil
uncut narwhal
so this:
```c
asm volatile ("movb %1, %%al;\
movb %%al, (...
asm volatile ("movb %1, (%0)" :: "r"(address), "{al}"(value) : "memory")
This is how inline ASM works
dense vigil
```
asm volatile ("movb %1, (%0)" :: "r"(address), "{al}"(value) : "memory")
```
Idk if GCC supports that syntax yet?
flint idol
wow thanks how fucking useful
uncut narwhal
Idk if GCC supports that syntax yet?
idk i use zig so im in clang land lmao
flint idol
wow thanks how fucking useful
(respondign to the code snippet from evalyn)
dense vigil
(respondign to the code snippet from evalyn)
Sorry
flint idol
its fine
uncut narwhal
Idk if clang has it either
that syntax is literally what im using
flint idol
clang doesn't like it
dense vigil
For legacy constraints
uncut narwhal
isnt capital A for the weird eax:ecx thing?
flint idol
asm volatile ("movb %%al, (%0)"\
::"r"(address),"A"((uint8_t)value)\
:"memory")```
this workedcool
dense vigil
flint idol
ill check with godbolt
dense vigil
Yeah it’s lowercase
dense vigil
isnt capital A for the weird eax:ecx thing?
Maybe?
uncut narwhal
yeah lowercase
uncut narwhal
capital A is for the eax:edx combo thingies but only on x86_32, it doesnt work on x86_64 ime
flint idol
MMIO_WRITE8(0x1000, 0);```
becomes this
```x86asm
mov edx, 4096
mov eax, 0
movb %al, (edx)```which looks about right
uncut narwhal
yeah that looks right
uncut narwhal
Maybe?
correction, capital A is for things that return a dword in ax:dx and not for things that return a qword in eax:edx lmao
blazing python
virgins use some mmio_write when real chads use volatile
uncut narwhal
Again who the fuck decided this syntax was good
gcc people back int he day™
uncut narwhal
virgins use some mmio_write when real chads use volatile
says the guy whose tag says sponsored by the brand that makes that not work
dense vigil
says the guy whose tag says sponsored by the brand that makes that not work
And only on checks notes bulldozer Microarchitecture CPUs for some fucking reason
Intel allows crazy shit on MMIO too
uncut narwhal
And only on *checks notes* bulldozer Microarchitecture CPUs for some fucking rea...
the newest bios and kernel developers guide from amd still lists the requirement so they could add it back at any time lmao
dense vigil
You can xor a register apparently and crap if you wanted
flint idol
# define MMIO_READ8(address)\
({\
uint8_t _result = 0;\
asm volatile ("movb %0, (%1)"\
:"=a"(_result):"r"(address)\
:"memory");\
(_result);\
// ...
uint8_t res = MMIO_READ8(0x1000);
})```
became this:
```x86asm
mov eax, 4096
movb al, (eax)```which also looks right
uncut narwhal
yep
dense vigil
the newest bios and kernel developers guide from amd still lists the requirement...
Can you get me the latest docs lmao
Somebody dosnt wana support this
uncut narwhal
you might be able to do some shit with the constraints to let it pick a more efficient™ memory addressing scheme but idk which ones are allowed for mmio and also idk the fuckin non-register contraints anyway im too stupid for that shit
uncut narwhal
Can you get me the latest docs lmao
i googled amd ppr and got this document from 2025 from them
dense vigil
What does configuration in this scenario mean?
real pecan
ECAM
uncut narwhal
yep
flint idol
fsdfsdpoi MMIO_WRITE32 doesn't compile
uncut narwhal
currently my kernel doesnt have generic mmio helpers, only ecam ones
flint idol
# define MMIO_WRITE32(address, value)\
asm volatile ("movl %%eax, (%0)"\
::"r"(address),"a"((uint32_t)(value))\
:"%eax","memory")```
"Asm-specifier for input or output variable conflicts with asm clobber list"
blah blah
just compile it
uncut narwhal
```c
# define MMIO_WRITE32(address, value)\
asm volatile ("movl %%eax, (%0...
youre still clobbering eax
at the end
dense vigil
```c
# define MMIO_WRITE32(address, value)\
asm volatile ("movl %%eax, (%0...
Your doing the eax thing there
uncut narwhal
What does configuration in this scenario mean?
this is the bottom of the page right before that
https://docs.amd.com/v/u/en-US/57930-A0-PUB_3.00 oh and heres the fuckin docs im looking at i should share that
flint idol
no PCIe on obos yet
uncut narwhal
ah you still only using the io ports? in that case yeah not yet
dense vigil
Why aren’t you using pice for xhci??
flint idol
do u even have pci
blazing python
do u even have pci
im waiting for iso day 1919
flint idol
giopsfdjs
blazing python
because some hw can forgive something other not
torpid wigeon
volatile loads/stores would maybe be ok if not for the fact that theres a brief ...
i really dont understand this ngl
flint idol
because some hw can forgive something other not
obos doesn't even boot on 99% of hardware
if it has a haswell cpu then it might
haswell is 2013 era
cpu
blazing python
haswell is 2013 era
i have laptop with celeron from 2012-2014 is it will run
flint idol
maybe
torpid wigeon
linux ignores this "problem" entirely btw
flint idol
ill send iso
flint idol
linux ignores this "problem" entirely btw
im not linux tho are i
torpid wigeon
then why do you need to care
flint idol
because dumb cuck amd says i do
dense vigil
And linux could very well have issues on certain weird ass hardware configs due to this problem technically
torpid wigeon
yet nobody is complaining so this seems like huge larp
and complexity that nobody cares about
blazing python
obos doesnt even boot on any amd hardware
fix ur kernel before doing xhci 🥀
dense vigil
I mean it’s still better than doing mmio->register and crap which is what OBOS does rn
dense vigil
Can’t that generate certain memory accesses not proper for registers?
Even if volatile
blazing python
I mean it’s still better than doing mmio->register and crap which is what OBOS d...
structs are more comfort than mmio_write() stuff
uncut narwhal
tbh for my kernel im doing mmio_write stuff more because i have a pathological distrust of the volatile keyword than because of this amd edge case that may or may not be real
its just that i can do like one letter different and never have to worry about whatever amd does again either in the process
torpid wigeon
volatile has some issues yes
but yes being able to read an entire struct with mmio at a time is kinda nice
uncut narwhal
i probably dont need to distrust volatile so much i just kinda do lmao
torpid wigeon
gcc has some details
uncut narwhal
yeah reading whole structs is def a point in favor of using mmio pointers
torpid wigeon
torpid wigeon
rip
flint idol
lmao
pi->usb3 = (((*dw0) >> 24) & 0xff) == 0x3;```faults here
hm
shrx what the fuck is that instruction gcc
"shift without affecting flags"
@torpid wigeon THIS is why you use inline assembly mmio macros
torpid wigeon
what's the problem
torpid wigeon
is dw0 volatile?
flint idol
i forgor
torpid wigeon
bro
uncut narwhal
bro...
blazing python
bro
flint idol
i think it's time to go
uncut narwhal
im on the pro-mmio-helpers side and i cant defend that lmao
blazing python
should boot™
how do you post 20 mb file without nitro
flint idol
ig server boost
torpid wigeon
@flint idol did volatile dw0 fix the codegen
uncut narwhal
lmao
flint idol
how do you post 20 mb file without nitro
wait can i send u an iso
with more logs enabled
blazing python
on my main pc it hangs there
flint idol
obos never booted on amd
never figured out why
probably never will, i dont own amd hardware
unless of course, patches sent
uncut narwhal
obos never booted on amd
what pte bits do you use for software purposes
flint idol
52 and 9
uncut narwhal
theres a bit in the pte thats ignored on intel and reserved-must-be-0 on amd thats actually checked for
flint idol
bit 9 is used for swap stuff, bit 52 is used to lock virtual memory (which is not done anywhere in the kernel, it only exists for the mlock and munlock syscalls)
uncut narwhal
bit 9 is used for swap stuff, bit 52 is used to lock virtual memory (which is no...
ok thats probably not it then, bit 8 is the one that got me
(page directories and up arent allowed to have the global bit set on amd)
flint idol
ic
uncut narwhal
only bottom level page entries are allowed to have that bit set
its ignored on intel so you can make your entire kernel tables have it set on their cpus
flint idol
i see
well i dont use it at all
as of today
(i might soon for optimization but idk)
uncut narwhal
so that isnt why obos dies on amd then lol
flint idol
nope
god knows why it dies on 99.9% of all x86 hardware
on my pc it gets a bit past lapic init
and hangs
on timer stuff
at first i thought that i should make an x2apic driver to fix it but that didn't do anything
blazing python
on my old laptop with celeron it just triple faults
flint idol
on my old laptop with celeron it just triple faults
oh well
that's sad
ill put it on my list of known broken hardware
(it is now 54983 entries long)
real pecan
i remember times when obos had proper suspend and resume
blazing python
obos prime is end
flint idol
short mortar
i need to fix my suspend and resume
uncut narwhal
i should do suspend/resume someday
flint idol
no
flint idol
i should do suspend/resume someday
no u shouldn't its acpi insanity
dense vigil
TIL obos has more rewrites than i do
pump thoes numbers up
flint idol
TIL obos has more rewrites than i do
the first four were all in 2023/2024
uncut narwhal
ive had one atm
torpid wigeon
c -> rust -> c (zinnia)
flint idol
no u shouldn't its acpi insanity
also theoretically my acpi suspend code is correct
for those who want to look at it for reference
uncut narwhal
neat
flint idol
its that something regressed somewhere else that causes suspend not to work properly
flint idol
also theoretically my acpi suspend code is correct
also tested on some real hardware and it worked
just the framebuffer does die on S3 suspend
on most hardware
flint idol
just the framebuffer does die on S3 suspend
cuz the gpu goes to sleep
if i get 60 stars today I will fix the bugs with obos booting on my pc
(5:30 hours until the offer is down)
uncut narwhal
stars on the thread?
astral mason
if i get 60 stars today I will fix the bugs with obos booting on my pc
dew it 
uncut narwhal
dang beat me to it, i just joined hte server on my alt to be the 60th lmao
but it wont let me react on the alt yet for some reason
apparently the 10 minutes before you can post also applies to reacting on threads
flint idol
dew it <:PichuKnife:1379169645084283012>
okay mark my words I will do this sometime in the next few days
Sunday at the latest
astral mason
I'll hold you to that >:3
flint idol
50 stars on the gh i will actually lock in for xhci
someone pin this
white mulch
50 stars on the gh i will actually lock in for xhci
flint idol
thx
re: my stupid syscall table code, I documented me doing it here lmao
wow how did i do this from 8 to 11 every day what
pretty much every obos native program doesnt use the native syscall api lol
flint idol
due to foreseen circumstances, I will have to delay fixing real hardware bugs by a few days
(I installed bmc4)
flint idol
guys
it is with great dismay
i announce the next rewrite of obos
because of its terrible instability
the new rewrite will be written in rust to aid in memory safety
and it will target aarch64 as it's main arch
and will run on primarily iphones as a replacement to ios
but will also provide full android binary compatibility
weak kestrel
It would have been believable if not for the last two lines
flint idol
@dense vigil
dense vigil
Why aarch64
flint idol
right
okay
the rewrite will also not support x86
and will not support any cisc arch
dense vigil
Will you do ppc32be
flint idol
maybe
that's the second port i will do
i will start the rewrite in a few hours, godspeed
inland radish
and will run on primarily iphones as a replacement to ios
it is my great pleasure to announce that i have smashed all my iphones
therefore i can't run your OS on any of them
oops! 😊
torpid wigeon
bruh
stark hamlet
6th rewrite incoming
flint idol
havent beat bmc4
but
tmpfs dev has been happening in the meantime
and this lol
VNODE_TYPE_CHR
switch (vn->vtype) {
case VNODE_TYPE_LNK:
*t = FILE_TYPE_SYMBOLIC_LINK;
break;
case VNODE_TYPE_REG:
*t = FILE_TYPE_REGULAR_FILE;
break;
case VNODE_TYPE_DIR:
*t = FILE_TYPE_DIRECTORY;
break;
default: OBOS_UNREACHABLE;
}
stats /root/test
create_vnode called on /dev/tty0
?????
uncut narwhal
flint idol
what the actual fuck
(what is "test")
(p is /root)
enum {
// This dirent hard links to the controlling tty of the process
DIRENT_REFERS_CTTY = BIT(0),
DIRENT_TMPFS_FILE_MOVED = BIT(0),
};```oh
uncut narwhal
well that dont look right
flint idol
ok u can now do this in the tmpfs thing:
mv ./test ..
mv ../test .```however u cannot:
mv test ..
nano test```
oh wait
fixed it
file removal also works
time to test xorg
nvm i have smth else to fix
it hangs when u want a login prompt
flint idol
dense vigil
Who needs thoes anyways
flint idol
hm interesting it writes to stderr but nothing comes out the other end
and read hangs
wait bash calls wait process then execs into /usr/sbin/logic
weird
what was it waiting on
oh
im dumb
i configured this build to put the login prompt on /dev/ttyS0
and forgot lol
xorg works with the new tmpfs
lol i just have some random binary called test in my initrd that accesses the network and i have absolutely no idea what it does
flint idol
First OBOS virus?
no
it ||is just a test of tcp keepalive that i wrote and forgot about||
flint idol
i committed tmpfs
dense vigil
Btw
flint idol
oh
flint idol
u can now mount devfs in different places
!!!
i have no clue what will happen if u say
write to the same block device at different paths
but let's just hope nothing funky happens
ok well since they have the same vnodes, it should work just like a hardlink
ok but since dirents aren't synced
if u do something like this:
cd ~
mkdir dev
mount -s devfs dev
cd dev
mkdir funy
ls
ls /dev/
rmdir funy
ls funy # nothing happens here except ENOENT
ls /dev/funy # it page faults here
which i actually only think will happen with directories
with files it just doesn't notice that the file is gone
and does random bs with the file that doesn't exist anymore
flint idol
i added some code to make it check if the file exists before returning the dirent
in the dirent lookup code
flint idol
which fixes this crash
but
the populate directory thing isnt smart enough to get rid of dirents that don't exist anymore
so just ls will still show deleted files
flint idol
the populate directory thing isnt smart enough to get rid of dirents that don't ...
i could just nuke the entry then repopulate it from scratch
that actually might be quicker than my other idea which was to mark all entries as FREED and if it's found while asking the fs driver for the dirents, the FREED marker is taken off
flint idol
i can't nuke each entry
in populate directory
because then entries that are directories will have children that refer to non existent parents
and memory bugs and blah blah blah
flint idol
that actually might be quicker than my other idea which was to mark all entries ...
this indeed works
flint idol
overlays work
u can do mount -o /dev/sdb1 /mnt and do whatever you want
and sync
and /dev/sdb1 will not be affected
uh nvm overlays may not work
lmao
i fixed that
flint idol
i hate when im hyperfixated and the surge of dopamine is in my brain then someone calls me and tells me to do something
jgljksdjhnfdas;lkfdsfasdqw
edwqrfokewjruesdhgltkendr;lhgtred
uncut narwhal
yeah that sort of thing fucks up the entire flow
flint idol
i hate when im hyperfixated and the surge of dopamine is in my brain then someon...
my poor PFC getting interrupted
(prefrontal cortex)
actually brain is very interesting
probably might study that thing after high school but who knows
flint idol
oh nice
unmounting doesn't work anymore
lmao
mounting a normal tmpfs also works
all i really have to do atp is to strip down the initrd driver's code to make it RO
like it was before i added the weird tmpfs bs to it about a year ago
and speaking of years, this obos rewrite is almost 2 years old
torpid wigeon
bruh
flint idol
i fixed it
diff --git a/src/oboskrnl/vfs/dirent.c b/src/oboskrnl/vfs/dirent.c
index 12363ba..e1837a6 100644
--- a/src/oboskrnl/vfs/dirent.c
+++ b/src/oboskrnl/vfs/dirent.c
@@ -103,7 +103,8 @@ static dirent* on_match(dirent** const curr_, dirent** const root, const char**
return cur_thr->proc->session->controlling_tty->ent;
}
- if (!only_cache)
+ bool vnode_is_drv = curr->vnode && (curr->vnode->vtype == VNODE_TYPE_DIR || curr->vnode->vtype == VNODE_TYPE_REG || curr->vnode->vtype == VNODE_TYPE_LNK);
+ if (!only_cache && vnode_is_drv)
{
dev_desc desc = 0;
driver_header* header = Vfs_GetVnodeDriver(curr->d_parent->vnode);
@@ -285,7 +286,7 @@ static dirent* lookup(const char* path, dirent* root_par, bool only_cache)
size_t currentPathLen = 0;
vdev* fs_driver = lastMount->fs_driver;
mount* mountpoint = lastMount;
- dirent* last = root_par;
+ dirent* last = Vfs_GetRoot() == root_par ? mountpoint->root : root_par;
while (tok < (path_mnt+path_mnt_len))
{
char* token = Vfs_Calloc(tok_len + 1, sizeof(char));
the fix btw
uncut narwhal
🎉
flint idol
ffs
noip is so annoying
they won't give me my dns record
remind me on the 22nd to get my hostname back
someone
kthx
flint idol
i made a side quest
and the ext2 driver now sets the right flags on mount and unmount
in the superblock
i eventually want to get /etc/mtab existing
i think ill either do some sort of sysfs-like thing for that
or ill get the mount and umount commands doing funny bs to get it to look like it works
i might look at fixing TCP selective ACK rn
as that is my only setback to merging tcp-congestion-ctrl iirc
if only i remembered
how the retransmission queue worked
whoever wrote this code was probably high on copium
tcp_sequence_rng_list rtxq;```
well it's just a list of TCP sequence rangesit would be a lot easier to fix the bug with SACK if i remembered what the bug was...
this man was insane
flint idol
im getting man-db to work
and i was able to get it to not hang anymore
but this
[ DEBUG ] (thread 60, process 48) syscall Sys_ExitCurrentProcess(0xff, 0x0, 0x0, 0x0, 0x0)```some process decided it wanted to be funny and exit with an error status
it calls getpid() then exit(-1)
because it thinks it's funny
oh it calls execve a few times before giving up on life
interesting
flint idol
oh it calls execve a few times before giving up on life
each time it returned OBOS_STATUS_NOT_FOUND
real pecan
why does exit current process have 4 damn arguments
flint idol
it doesn't
the syscall logger is too lazy to know what arguments actually exist
so it prints all 5 registers
real pecan
lol
flint idol
OBOS_Debug("(thread %ld, process %ld) syscall %s(0x%p, 0x%p, 0x%p, 0x%p, 0x%p)\n", Core_GetCurrentThread()->tid, Core_GetCurrentThread()->proc->pid, syscall_to_string[eax], rdi,rsi,rdx,r8,r9);
oh man-db wants /bin/sh
for nroff
well it does indeed have /bin/sh
but the leading space caused my execve implementation to mess up
#! /bin/sh```(it still doesn't work)
i swear man-db is the most retarted program ever developed
how tf do i run gcc but not man-db
some stupid fuckass program in the pipeline that man-db starts
is supposed to convert the man page to smth readable
but NOOO
it outputs nothing to the pager
not even
it doesn't even write to a pipe
with anything
i think nroff is the culprit
groff is actually fucking retarted what
"prevent destructor from recursing"
wtf
how tf is fflush returning ENOENT???
real pecan
obos stability strikes again
weak kestrel
lmao
flint idol
Sys_FdFlush
isn't
even called anywhere
😭
mlibc seems to be hallucinating errors from fflush
although fflush might not call fsync now that i think about it
the EC im getting from groff
is a bug
ENOENT is from an old errno value
flint idol
is a bug
as fflush does not set errno iirc
or does it wait what?
fflush doesn't seem to set errno in mlibc unless im mistaken
right so i love how
fd 1 in some process started by nroff
is a pipe
but there is no handle clone/fcntl syscall
changing stdout to a pipe
dxscfsdfs
i can run xorg and gcc but not fucking groff the game is rigged against me sdafjsdklpfjdms
ok so far i see that it does this:
pipe(&fds);
fork();
fork();```and that's the only time a pipe is even created
how is fd 1 a pipe
[ DEBUG ] (thread 58, process 46) syscall Sys_IsATTY(0x1, 0x0, 0x0, 0x0, 0x0)
[ DEBUG ] (thread 58, process 46) syscall Sys_IsATTY returned 0x27 (OBOS_STATUS_NOT_A_TTY)
[ DEBUG ] (thread 58, process 46) syscall Sys_FdSeek(0x1, 0x0, 0x1, 0x0, 0x0)
[ LOG ] (thread 58, process 46) syscall Sys_FdSeek returned 0x2f (OBOS_STATUS_SEEK_UNALLOWED)
[ DEBUG ] (thread 58, process 46) syscall Sys_FdTellOff(0x1, 0x0, 0x0, 0x0, 0x0)
[ DEBUG ] (thread 58, process 46) syscall Sys_FdTellOff returned 0x0 (OBOS_STATUS_SUCCESS)
[ DEBUG ] (thread 58, process 46) syscall Sys_FdWrite(0x1, 0xef4008, 0x97, 0x8b6588, 0x0)
[ DEBUG ] pipe.c:write_sync: closed pipe
[ LOG ] (thread 58, process 46) syscall Sys_FdWrite returned 0x21 (OBOS_STATUS_PIPE_CLOSED)
fd one is a pi
pe
yet
a pipe is never put on fd 1
im gonna stop repeating myself and debug ts
ill just ignore that
i believe the 2nd child of groff owns the reading end
only problem is uh
that process dies before the write ever happens
all that process does is:
- start
- execve
- check if fd zero is a pipe
- check if fd one is a pipe
- exit
wait
it reads from fd zero
but the read is successful
despite the pipe being empty
flint idol
i think i fixed pipes finally
cuz i got groff to work
i spoke too soon
nroff works but noooo man-db doesn't
flint idol
syscalls have 5 parameters yes
inland radish
i'm out here with 9 parameters max
flint idol
damn
inland radish
(though on i386 i have to work around the lack of gprs so i end up doing some extra stuff)
short mortar
(though on i386 i have to work around the lack of gprs so i end up doing some ex...
I just overflow to stack
flint idol
has that ever been proven useful
inland radish
I just overflow to stack
i do have to do that with i386
flint idol
if i need more than 5 arguments i just have the last one a pointer to a struct with the extras
short mortar
I use uint64_t for everything, so all of my syscalls are super long on ia32
inland radish
has that ever been proven useful
i have a function that takes 7/8 arguments
short mortar
But I also have syscalls that take like 10 uint64_t's
inland radish
(8 on 32-bit because there's one 64-bit argument)
behold:```c
BSTATUS OSWriteFile(PIO_STATUS_BLOCK Iosb, HANDLE Handle, uint64_t ByteOffset, const void* Buffer, size_t Length, uint32_t Flags, uint64_t* OutSize);
Nevermind, it's 8 parameters
But I just pass a struct to kernel because it's too much
inland radish
runners up:
BSTATUS OSMapViewOfObject(
HANDLE ProcessHandle,
HANDLE MappedObject,
void** BaseAddressInOut,
size_t ViewSize,
int AllocationType,
uint64_t SectionOffset,
int Protection
); // 7 parms
BSTATUS OSCreateSymbolicLink(
PHANDLE OutFileHandle,
HANDLE DirectoryHandle,
const char* FileName,
size_t FileNameLength,
const char* TargetName,
size_t TargetNameLength
); // 6 parms
BSTATUS OSCreateThread(
PHANDLE OutHandle,
HANDLE ProcessHandle,
POBJECT_ATTRIBUTES ObjectAttributes,
PKTHREAD_START ThreadStart,
void* ThreadContext,
bool CreateSuspended
); // 6 parms
So 64 bytes
weary hound
i think in managarm it's at most 7 args (6 in + 1 out), for helCreateThread and helMapMemory ```c
HelError helMapMemory(
HelHandle handle,
HelHandle space,
void *pointer,
uintptr_t offset,
size_t size,
uint32_t flags,
void **actual_pointer);
HelError helCreateThread(
HelHandle universe,
HelHandle address_space,
HelAbi abi,
void *ip,
void *sp,
uint32_t flags,
HelHandle *handle);
interesting
my biggest syscall is probably mmap
it has 5 physical parameters but one is a pointer to a struct with 3 fields as 3 extra parameters
real pecan
i think in managarm it's at most 7 args (6 in + 1 out), for `helCreateThread` an...
What's a universe
weary hound
a space within which handles exist
threads have an universe attached in which handles used by that thread are resolved, and the universe can also be manipulated by the holder of the handle to it
this is used by e.g. posix-subsystem to give file stream handles to processes, which on open inserts a handle to the stream into a process' universe, then writes the new handle into shared memory that maps fd->handle
weary hound
currently always kHelAbiSystemV (1) :^)
weary hound
yeah the kernel doesn't really make much use of it
in the future ig one use could be if we want to provide linux syscall emulation or whatever, the abi could be used to disable native syscalls and make all syscalls become observations
real pecan
I see
real pecan
in the future ig one use could be if we want to provide linux syscall emulation ...
Ah
So its more like personality
weary hound
i guess, but nothing's been decided yet
real pecan
Fair
weary hound
threads have an universe attached in which handles used by that thread are resol...
in essence, handles in and of themselves don't mean anything, but (universe, handle) references some specific kernel object
for most syscalls it's implied that it uses the thread's universe
and there are syscalls to manipulate an universe by handle to it
real pecan
can obos survive a fork bomb
flint idol
it hangs
technically it doesn't hang
but it is rendered inoperable
as there is nothing u can do to kill the fork bomb
libpipeline is so stupid i hate libpipeline
flint idol
it also hangs
does the kernel itself deadlock or is it just rendered inoperable cuz of all the processes
weak kestrel
I'm not sure
real pecan
as there is nothing u can do to kill the fork bomb
Does it not oom
flint idol
i dont think so
it could've OOMed, or the kernel could be aggressively swapping and i have no idea
man-db is so dumb what
for some reason
zsoelim and manconv
completely nuke the input manpage
which is one of the reasons why man-db isn't working
knotty totem
mandoc ❤️
flint idol
atp i might as well port mandoc instead
flint idol
todo fix more bugs today
flint idol
i don't use ai for obos at all
but i'm keeping it here because i can't be damned to move it
uncut narwhal
mood
tbh if i didnt want to reorganize my top post and add a pinboard message i wouldnt have bothered for mine either
flint idol
man ls.1 doesnt work but who cares!!
i'm gonna look for a reprod on linux mlibc
flint idol
as expected, this does not happen on linux mlibc
blazing python
i thought obos was like vibos
short mortar
vibes
Lol, what if I start a vibecoded os in some cursed language or something as a second project?
(though now that I think of it, it's a bar idea)
flint idol
OBOS (not vibecoded)
flint idol
maybe debugging today?
blazing python
maybe debugging today?
xhci
honest citrus
maybe debugging today?
PowerPC port
weak kestrel
zig rewrite
blazing python
rust rewrite
short mortar
c++ rewrite
torpid wigeon
c rewrite
real pecan
ada rewrite
blazing python
v rewrite
uncut narwhal
lol
uncut narwhal
maybe debugging today?
debugging.
real pecan
Rust rewrite day
uncut narwhal
lmao
uncut narwhal
no debugging today
F
weak kestrel
porting systemd day
dense vigil
Taking a break and not overworking yourself day?
torpid wigeon
whne c rewrite
blazing python
when hoi4 port
real pecan
Jai when
flint idol
Taking a break and not overworking yourself day?
no more like overworked by school and not feeling up to debugging mandoc
when there is much easier forms of dopamine to be found
uncut narwhal
the nice thing with hobby projects is theres no deadline lol
flint idol
real lol
flint idol
when there is much easier forms of dopamine to be found
for example the dopamine of finding a new project to work on that u know u don't have the motivation to exactly start but definitely to explore
flint idol
i have to have motivation for this project considering ive already dropped 20 dollars on it
fucking canada post is a scam
15 dollars shipping for a 5 gram package
real pecan
what did u buy
flint idol
intel 8086
flint idol
i have to have motivation for this project considering ive already dropped 20 do...
new project*
flint idol
intel 8086
the actual thing is $5 dsfsdfds
(for reference i can ship in two ICs from china for $8)
lean glen
whats annoying with the 8086 and stuff is you have to find a bunch of components and they're not getting made anymore
I've found sellers on ebay selling a bunch of pc components
but its annoying
not like the 6502
lean glen
I have designed a few computers but never built one
flint idol
with 8086?
lean glen
itd be nice to do it but it's easily like 200$
lean glen
with 8086?
8088
flint idol
ic
uncut narwhal
i wish zig had 6502 support tbh. (or actually 65c816 tbh), id design up a machine for that happily to play with
flint idol
m68k (/j)
lean glen
i wish zig had 6502 support tbh. (or actually 65c816 tbh), id design up a machin...
llvm has one
flint idol
ill probably start the design of the 8088 thing rn
uncut narwhal
llvm has one
not officially part of llvm proper so i cant use it with zig really
(side note that someone managed to snipe issue #6502 for the git issue tracking 6502 support for zig and i find that hilarious)
lean glen
ill probably start the design of the 8088 thing rn
depends on what you want to do
do you wanna build a PC compatible?
if so you'll need a bunch of components
id rather do something like 6502 which is boring but its readily available
weary hound
very topical, i just ordered a pcbs from jlcpcb for my 65816 project :^)
flint idol
i just need a side project that will keep occupied for a while even if im not actively doing it
weary hound
but yours is a fpga right
yeah the pcb is just a 40 pin header to dip adapter
plus passives
weary hound
but yours is a fpga right
to be specific, it's an actual 65816, with the fpga being the rest of the owl
weary hound
yeah i don't see a reason why this wouldn't be possible with other cpus
lean glen
yeah i don't see a reason why this wouldn't be possible with other cpus
there were commercial PC chips even
idr the name
faraday
real pecan
but im really bored
What did you want to do with it?
real pecan
Like eat it, destroy it, port obos to it or
lean glen
I dont think its really feasible to port an OS to a breadboard computer
because as soon as you add an mmu it gets very messy
flint idol
Like eat it, destroy it, port obos to it or
eat it
prob just write some code to drive it
knotty totem
because as soon as you add an mmu it gets very messy
You can run a 286 on a breadboard so...
I think there's even been a 3 or 486 project maybe?
lean glen
You can run a 286 on a breadboard so...
you need a breakout board
which is not cheap nor readily available
knotty totem
which is not cheap nor readily available
I mean it's not super cheap but it is doable
There might be cheaper ones on aliexpress, but last I tried looking it was so hard to find anything actually related to what I want 🙄
flint idol
elfin belfry
isn't this the standard tho?
flint idol
id have preferred if one of them were active high for simplicity's sake
as now i need both a RD and WR signal on the bus
elfin belfry
well, flash memory ICs have them active low too
flint idol
although maybe that's standard of most chips
it seems like i'd have needed such a signal anyways
for my rom
whatever
it's probably not that deep
lean glen
id have preferred if one of them were active high for simplicity's sake
active high is not standard
flint idol
it's probably not that deep
this
flint idol
i wonder whether i should disable EEPROM writes in my bus logic
and by that i mean just tying #WE high on the EEPROM
instead of connecting it to the bus #WR signal
elfin belfry
i don't think you should use eeprom as write mem on your pc
but use a flash instead
flint idol
i see, any particular reason?
elfin belfry
eeprom doesn't like being re-writted that often
it will prob die quite fast compared to flash
flint idol
okay
weary hound
instead of connecting it to the bus #WR signal
could maybe wire it up through some gate and some ff to allow enabling writing
WE# = or(WR#, ~enable FF)
then have the FF be writable through a bit on some io port or w/e
flint idol
that also would work
but i like the io port idea better
i'm drawing out the bus logic on a piece of paper rn and i'm gonna put it in my schematics later
i'm also gonna have to make sure my clock logic is sound
as rn it's just a passive oscillator connected to an 8284
and the output of the 8284 is connected to the 8088
weary hound
are you gonna use any programmable logic or just generic logic chips
flint idol
are you gonna use any programmable logic or just generic logic chips
just generic logic chips
weary hound
ah
flint idol
i've completed the memory space address decoding logic
it wasn't too difficult, address lines 16-19 are NANDed to produce CS for ROM, and inversed for the RAM CS
flint idol
and then there is some more logic in the ram because i am using two 32K RAM chips to address 64K of ram
flint idol
and then there is some more logic in the ram because i am using two 32K RAM chip...
just some logic to choose either the first or second chip based off A15
this is what i came up with on paper
for the ram decoding thing
not the prettiest sketch
but i've drawn worse
uncut narwhal
most important is can you read it lol
flint idol
yea lol
i've implemented this in my drawing
but i realized rn that in my schematic
i completely disregard the IO/M pins
pin*
which means that all bus accesses are treated as memory accesses
in my sketch the logic exists
but not in my schematic so i'm gonna have to fix that
flint idol
but not in my schematic so i'm gonna have to fix that
nvm
i did put it in the schematic
flint idol
for today i will be drawing out the logic for the I/O ports
i think i will just have address lines 15 thru 8 determine the device selected
with each bit acting like a flag
oh right this is obos thread lol
whatever
obos port to 16 bit arch
i also wrote this silly little rom
for when/if i build the thing
oh come on i'm gonna need a flash programmer
damn it
where am i supposed to get that
oh wait amazon
dense vigil
oh come on i'm gonna need a flash programmer
You can also use an arduino or pi pico
If it’s a DIP chip and a parallel flash rom
It’s really easy ish
flint idol
You can also use an arduino or pi pico
that's what i'm gonna do
flint idol
i just wrote a flash programmer for the ROM i'm gonna use